What can be done with a data center’s waste heat

A bakery that warms the flat upstairs is not doing it on purpose. The ovens run, the heat rises, and the neighbours notice. If the baker ran a pipe to the flat, that warmth could be put to work.

Data centers produce heat all day, every day. This piece looks at what that heat can be used for, where reuse works well, and why in many places it still goes straight into the sky.

Why there is so much heat

Almost all the electricity a data center uses ends up as heat. (We explain why AI draws so much power in Why AI uses so much electricity, in plain words.) AI sites use a lot of power, so they produce a lot of warmth.

Normally, that heat is pushed outside through cooling towers or dry coolers. It is treated as a waste product. Reusing it means capturing it before it leaves and sending it somewhere useful.

Not all heat is equal

The key question is temperature. Warm air or water that is only a little above room temperature is hard to use. Hotter water is much easier.

Air-cooled sites tend to produce low-grade warm air. Liquid-cooled sites can produce warmer water, because the liquid picks up heat right at the chip. (See Air cooling or liquid cooling for AI?) That makes liquid cooling a better starting point for reuse.

Even so, data center heat is usually cooler than what most heating systems expect. Heat pumps are often used to raise it to a useful temperature. They use some electricity to do that, which is worth counting.

Where it can go

District heating. Some towns run hot water through pipes to many buildings. A data center can feed heat into that network, helping warm homes, offices or schools.

Single buildings. A nearby office, pool, hospital or apartment block can take heat directly, if it is close enough and its heating system can use it.

Greenhouses. Plants need steady warmth, especially in cooler months. Greenhouses near data centers are a natural fit.

Fish and food. Fish farms and some food processing need warm water or warm air at moderate temperatures.

Industry. Some processes need low or moderate heat, such as drying materials. Others need far hotter heat than a data center can provide.

Where it works

Reuse tends to work when several things line up.

Someone nearby needs heat. Heat does not travel far without losses and cost. A user close to the site matters more than almost anything else.

The need matches the supply. A data center makes heat all year. Users who need heat all year, like greenhouses or some industry, fit well. Home heating is seasonal, so summer heat may still have to be rejected.

The infrastructure exists, or can be built. An existing district heating network makes reuse much easier. Building new pipes is possible but slow and costly.

Agreements are clear. Someone has to own the pipes, pay for the heat pumps, and agree what happens if either side stops.

Where it does not

In many places, reuse is hard to justify.

Warm climates need little heating. Remote sites, often chosen for cheap power, may have no neighbours who need heat. Small sites may not produce enough heat to make a connection worthwhile. And the site still needs a way to dump heat when the user does not want it.

Rules and incentives vary by place. This article is not legal advice, so check local rules for any specific site.

What it means for siting

If heat reuse matters to you, it has to be planned early. It affects where the site goes, how it is cooled, and who it partners with locally.

That can pull in two directions. Capacity is moving toward available power. (See Why AI capacity is moving toward the power.) Heat users are often in towns. Smaller, modular sites can help here, because they can be placed nearer to both. (See Edge data center: can a small town host AI capacity?)

Questions to ask

If heat reuse is part of your plans, these questions help.

1. What temperature is the heat, and is it from air or liquid?
2. Who nearby needs heat, and in which months?
3. Is there a heating network, or would pipes need to be built?
4. Are heat pumps needed, and how much power do they use?
5. What happens to the heat when the user does not need it?
6. Who owns and maintains the connection?

What we are building

We look at heat as part of choosing and designing a site, not as an afterthought.

We're building standardized compute close to power, connectivity and demand, joined by one programmable network for AI inference. The network is designed to place workloads by policy, on latency, data residency, model and load, and every workload is designed to generate an auditable record of commitment, placement and delivery.

Heat reuse is not right for every site, but where it fits, it is worth planning from day one.

If you own a site or a heat network and want to talk about hosting AI capacity, use Request capacity below.